Running hypertable inside docker

[Teodor Pripoae]

Hi,

I'm trying to run Hypertable on our Docker/Kubernetes setup. I have a few questions:

1. I will use Ceph broker since Kubernetes also has support for it. I'm planning to completely detach the DFS from the distributed cluster so all other services can be immutable storing the state on the DFS. Does Hypertable have a performance requirement for FsBroker to be on the same nodes as Ceph/Hadoop volumes ? I'm not planning to run MapReduce at the moment. I will run it inside a 3 zone region on Google Compute Engine, so latency should be like "rack latency" inside one zone and "datacenter lantency" between zones.

2. I can see in the hadoop tutorial that Hypertable should be configured using "ht cluster" command. I'm planning to setup docker images for each of the 5 services (FsBroker, Master, Hyperspace, RangeServer and ThriftBroker). Is there any way to generate a static configuration for each service ? If I put the same cluster.def inside each container will it work by default without running ssh or ht cluster manually ?

3. I can see from the documentation that Hyperspace failure is not tolerated and needs a full cluster restart. Which systems directly connect to Hyperspace ? From the graphic present on Architecture page it seems only ThriftBroker connect to Hyperspace. Is this true ? 

4. In case there is a network partition between Hyperspace and services depending on it, after the partition resolves, will everything work without manual intervention ?

5. If Hyperspace is restarted, does an operator need to do a full cluster restart ? Kubernetes assigns a static IP for each service and disk volume can be stored on a DFS, so in case of machine failure Hyperspace can be moved transparently to another server with only a few seconds of downtime. What is the behaviour of services depending on it in case there is a 10-20 downtime due to Hyperspace restart ? 

6. Does Hyperspace store any in memory state which needs cluster restart ?

Thank you,

Teodor